Before ASB
Manual every time- Repeat the same clicks and copy-paste every run
- A human has to stay in the loop
- Handoffs break when context is missing
- Difficult to reuse across assistants and teammates
Local-first AI automation platform
visualbuild.me · Automation Skill Builder
Record once. Run forever — no AI required at runtime.
Use AI once to capture the workflow, then compile it into deterministic runtime logic you can rerun locally without AI in the loop.
- Record with AI, run without AI
- Deterministic local runtime
- Works with Cursor, Claude & Copilot
- Reusable as standard MCP tools
HOW IT WORKS
Connect → Capture → Deliver
Local-first — not a cloud sandbox agent. AI is used once at record time; zero cloud calls at runtime.
- Desktop automation
- Browser automation
- MCP tool surface
- Skill packaging
WHY IT CLICKS
From repetitive manual work to reusable runtime logic
No runtime AI required
After ASB
Reusable runtime logic- Record the workflow once with AI assistance
- Compile it into deterministic runtime logic
- Run it again locally in one click
PROMPT EXAMPLES
Tell the AI what to do — in plain language
Paste a prompt into your MCP client. ASB executes locally — browser, desktop, ERP — and reports each step back.
amazon.com
Open amazon.com, search "iPhone 17", extract product names and prices from the results page, and save them as iphone_prices.csv on the desktop.
EXECUTING
- Opened amazon.com
- Searched "iPhone 17"
- Extracted 12 product rows
- Wrote iphone_prices.csv to Desktop
MCP · ASB · AI design
Use MCP ASB to enable AI design mode, record desktop steps via MCP tools (not WinMCP), save as pass_action.py, run AI parameter analysis, and package as exe.
Use MCP asb to enable AI design mode and start AI design recording. Then use tools: wait 2 seconds, click 1406,257, wait 2 seconds, input ctrl+a, wait 2 seconds, set DGJW26010091 to the clipboard, then Ctrl+v to paste, wait 4 seconds, click 492, 373, wait 3 seconds, click 410,933, wait 3 seconds, click 635,270, wait 1 second, use ctrl+a to select all, wait 1 second, set test to the clipboard then Ctrl+v, wait 2s, click 1320,526. Execute all via MCP calls; do not use winmcp tools for mouse/keyboard operations. After recording ends, stop recording and save as pass_action.py. Use the MCP tool AI to analyze parameters: input parameter code with value DGJW26010091, parameter tip with value test, then package as exe. Remember to use the AI tools in MCP, not yourself.
EXECUTING
- mcp: enable AI design · start_recording
- wait 2s → click (1406, 257)
- Ctrl+A → clipboard DGJW26010091 → paste
- click (492,373) → (410,933) → (635,270) …
- stop_recording → pass_action.py
- mcp: ai_analyze_params (code, tip)
- mcp: package_exe
164_sap_sap · SAP GUI
Use MCP ASB tools starting with 164_sap_sap — prefer an existing connection (create one only if none exists). Open transaction ZVF04 to maintain the invoice due-date list, wait 5s, then enter sales organization 1010.
EXECUTING
- 164_sap_sap_list_connections
- 164_sap_sap_open · existing connection
- Transaction ZVF04 — invoice due list
- wait 5s
- input sales org 1010
SUBSCRIPTION LICENSE
Simple, honest pricing
Each time an AI assistant invokes a skill via MCP, that counts as 1 call. A typical automation run uses around 20–30 calls.
Start free — no credit card needed. Every account gets 1,000 calls per month at no cost, resetting on the 1st of each month (your local time). When you're ready to do more, upgrade any time.
FREE
$0
1,000 calls / month
≈ 33–50 skill runs / month
Get startedSTARTER
$9 / mo
5,000 calls / month
≈ 165–250 skill runs / month
PRO
$29 / mo
20,000 calls / month
≈ 665–1,000 skill runs / month
GROWTH
$99 / mo
100,000 calls / month
≈ 3,300–5,000 skill runs / month
ENTERPRISE
Custom
Volume · SLA · deployment & integrations
Contact usTypical run: 20–30 calls. A skill with loops or retries may use more.
Add-on: +1,000 calls one-time for $2 · All prices USD before tax
Questions & answers
Short answers for common pre-sales and install questions.
What are the system requirements?
Summarized for operators and procurement. Pre-built desktop bundles target Windows and macOS (executable + supporting files).
Baseline (bundled app): The ZIP includes what you need to start the service—no separate Python install is required as a precondition. Use a 64-bit OS, reserve 8 GB+ RAM for heavier OCR / ML / desktop workloads, allow several GB of disk for caches on first runs, and use a current browser for the local web UI (often http://127.0.0.1:8800). Optional Python or Node.js may be needed for specific skills.
Windows
64-bit Windows 10 or later (or supported Server). Extract the ZIP and launch the packaged executable; allow prompts for permissions as documented.
WSL counts as Linux for dependencies, not Windows-native UI automation. Optional runtimes (Python, Node.js) may be required for specific skills.
macOS
Use a recent macOS release that runs the distributed bundle. Extract the ZIP and open the app; approve Accessibility / Input Monitoring when the system prompts.
Optional runtimes (Python, Node.js) may be required for specific skills; browser automation also needs MCP configuration in the app.
Linux
The Linux edition suits Debian-based environments. Because deliverables can be large, please reach out to us (or your maintainer) for details on distribution and how to obtain the build.
Headless or server deployments may require DISPLAY or --no-gui.
Is this a hosted SaaS?
No. It is designed primarily as a self-hosted application you run on your infrastructure or
workstation. You control networking, TLS, and who can reach the service.
Do I need MCP or Claude community skills?
No. Core features work over the web UI and HTTP API. MCP and integration with the rich skill ecosystem on the Claude community are optional when you want assistants or a skill runtime to consume packaged capabilities. We fully support that path: use the in-app skill manager to install the skills you prefer, use the product’s built-in security checks to validate them, and execute with confidence.
Will this replace my team?
No — it amplifies what your team can do. Think of it as a capable assistant that drafts and suggests; your
team reviews, decides, and owns the outcome.
What about data privacy?
Sensitive data flows depend on your configuration: local execution, which models you call, and
which external services you enable. Run data-protection assessments before processing regulated data.
With this product’s design, interactions are auditable and can be reviewed in a visual way—helping security teams verify behavior instead of working blind.
Why local-first?
Many automation tasks touch the desktop, filesystem, or internal APIs. Keeping execution close to the
environment reduces latency and can simplify compliance—at the cost of you maintaining the runtime.
Does execution call the cloud?
At runtime, zero cloud or AI calls for skill execution — invocation, quota, and license checks stay on your machine. AI is used when you record / author (with the model you configure), not on every rerun.
Paid tiers still need a periodic license.json refresh (~every 30 days) from our site. That is a one-time file update, not a per-run network dependency.
What happens if I stay offline past my license expiry?
- If your license is still valid (
expires_athasn't passed), you can run completely offline indefinitely until it expires. - If
expires_atpasses while offline, the app automatically falls back to the Free tier (1,000 calls/month) until you reconnect and refresh your license. - If you cancel your subscription but your license file hasn't expired yet, you'll continue running at your paid tier locally until expiry — this is a known trade-off of the offline-first design.
Note: like most offline license systems, this relies on your system clock being roughly accurate. We don't currently implement clock-tamper detection — this is a deliberate trade-off in favor of true offline operation.
Will VS Code, Chrome, or Office integrations be available?
Yes—see the
Visual Studio Marketplace
listing or download a VSIX. The Automation Skill Builder app must still run locally;
extensions only embed the UI. A Microsoft Office add-in remains under evaluation.
What if OCR or vision is slow the first time?
Vision stacks often download models on first use. Expect a longer cold start; subsequent runs are typically
faster. Warmup and skip flags are available in the app settings.
How do I get help?
Community users can ask questions in the public forum — response times vary. If your team needs reliable
support with a guaranteed response, check out our professional plan.
Do I need Python or Node.js installed?
No. The bundled app includes what you need to start the service—no separate Python install is required as a precondition. Optional Python or Node.js may be needed for specific skills. Pre-built bundles target 64-bit Windows 10+ and recent macOS.
Support model
Two lanes—so expectations stay clear whether you are evaluating, self-hosting, or running under a commercial agreement.
Community & as-is use
The software may be used and inspected according to its license and repository terms. Documentation, issues, and community discussion (where offered) are provided without warranty.
Community support is best-effort — driven by fellow users and the maintainer when possible. It’s ideal for exploring, learning, and running personal or internal projects. For teams that need faster answers or hands-on help, professional support is available separately.
Professional support
For organizations that need onboarding, priority help, custom integrations, security review assistance, or deployment hardening, commercial support may be available from the maintainer or partners.
Scope, channels, and SLAs are defined in a separate agreement—not on this marketing page. Contact your vendor or project owner for entitlement and pricing.
Contact
Billing and subscriptions: info@visualbuild.me
Technical and product issues: dev@visualbuild.me